HAI’s Background:
Stanford’s Institute for Human-Centered AI (HAI) was established in 2018-2019 to address the rapid growth of AI and its potential impact on society. AI has evolved from a niche computer science discipline to a transformative technology driving the fourth industrial revolution.

HAI’s Mission:
HAI emphasizes the importance of considering the human perspective and societal impact alongside technological advancements in AI. The institute aims to advance AI research, education, policy, and outreach to improve human conditions and address potential unintended consequences.

Research Focus Areas:
HAI’s research explores various aspects of AI’s impact on society, including: AI ethics and responsible AI development. Human-AI interaction and collaboration. AI’s impact on work, education, and healthcare. AI’s implications for social justice, equity, and inclusion. AI policy and governance.

Fei-Fei Li’s Human-Centered AI Research:
Fei-Fei Li and her team’s research is guided by three principles: AI is interdisciplinary and requires collaboration with social scientists and humanists. AI should augment human capabilities and enhance humanity instead of replacing them. Human-inspired AI should be flexible, robust, and able to learn with limited labeled data.

ImageNet Dataset and Deep Learning Revolution:
Fei-Fei Li co-developed the ImageNet dataset in 2006 to establish a benchmark for object recognition. The goal was to create a large-scale dataset to push the field of computer vision forward and seek the North Star of visual intelligence. ImageNet inspired a new approach to object recognition by leveraging big data and deep learning algorithms. The dataset’s creation involved a leap of faith as the memory and chip capacities at the time were limited. The success of ImageNet led to a surge in deep learning research and contributed to the deep learning revolution.

The Significance of ImageNet:
ImageNet’s importance lies in enabling data-intensive deep learning methods to materialize and be researched. It provided a large-scale dataset for benchmarking object recognition, which was a crucial problem in computer vision and machine learning.

Fei-Fei Li’s Perspective:
Li highlights the belief in a guiding principle, the “North Star,” which was the problem of object recognition. She emphasizes the importance of establishing this problem as a central focus for research in the field. Li acknowledges that ImageNet’s success was built upon generations of prior work and collaborative efforts.

ImageNet as a Benchmark:
ImageNet served as a standard for benchmarking progress in large-scale object recognition. It provided a common platform for researchers to compare and evaluate their approaches. The availability of a large and diverse dataset enabled the development of more accurate and robust deep learning models.

Conclusion:
ImageNet’s role in advancing AI and deep learning is undeniable. It facilitated the development of data-driven approaches, leading to significant progress in object recognition and computer vision. The establishment of a shared benchmark played a crucial role in accelerating research and promoting collaboration within the field.

ImageNet’s Impact on AI:
Fei-Fei Li highlights the importance of ImageNet in revitalizing the field of neural networks. Before ImageNet, machine learning algorithms were smaller scale and required manual parameter tweaking. ImageNet enabled the use of high-capacity models like neural networks due to its vast data and powerful computing chips.

Data’s Significance in AI Development:
Fei-Fei Li emphasizes the crucial role of data in the advancement of AI, particularly in the past decade. Alexandr Wang raises the question of how much investment organizations should make in developing data sets.

Data as a Means to an End:
Fei-Fei Li views data sets as a means to achieve scientific goals, rather than an end in themselves. Her scientific quest focuses on solving important problems in visual intelligence and AI. Fei-Fei Li believes that data sets should be created with specific scientific goals in mind.

Additional Investments in Data:
Fei-Fei Li suggests that organizations should invest in data sets that align with their scientific goals. She emphasizes the need for data sets that are diverse, representative, and relevant to the problem being addressed. Fei-Fei Li advocates for investing in data collection, annotation, and curation to ensure high-quality data sets.

Data Sets and Their Role in AI:
Data sets play a crucial role in AI by defining problems and providing fuel for algorithms. In the business world, data sets are tools used to meet customer needs, especially human-centered ones. Unintended consequences, such as bias, may arise from data sets, so it’s essential to mitigate these issues.

Mitigating Bias in AI:
Raising awareness about the unintended consequences of technology, particularly bias in data and algorithms, is crucial. Embedding ethics education into computer science curricula to create a generation of tech-savvy and ethically-minded technologists. Investing in algorithm and data development technologies to de-bias or avoid bias. Advocating for interdisciplinary inclusion of ethicists and philosophers in algorithm design. Establishing governance mechanisms for data and algorithm oversight, including legal and policy frameworks.

North Stars in AI Research:
AI research has become diverse and offers many opportunities, making it challenging to identify specific North Stars. Large language models in natural language processing have made remarkable progress, highlighting the field’s dynamism. Embodied intelligence and immersion in the visual world are becoming increasingly important areas of research in computer vision. Understanding intelligence as a tool for survival, seeking food, avoiding predators, communicating, and socializing is crucial for advancing AI research.

Fei-Fei Li’s Research Interests:
Transition from passive visual intelligence to an embodied and active approach. Excited by the interaction between robotics and vision. Research focus on embodied agents, combining visual intelligence with planning, learning, and studying how complex agents emerge from real-world interactions.

Evolution of Human Feedback in Algorithm Learning:
Current data annotation is static and third-party. Humans learn dynamically from other humans.

Human-in-the-Loop and Collaborative Learning:
Human-in-the-loop and collaborative learning are gaining attention. Recent project in Fei-Fei Li’s lab explored engaging humans in loop to improve algorithm performance. Human feedback can be used to correct mistakes and refine models. Collaborative learning can improve algorithm performance and make it more adaptable.

Active Learning and Human Collaboration:
Fei-Fei Li emphasizes the importance of active learning in AI, where the AI agent proactively seeks meaningful conversations with humans to learn visual concepts. This approach allows the agent to learn what it wants to learn and engages the human in a win-win collaboration. Active learning research includes reinforcement learning algorithms with engagement rewards and online learning techniques.

Cognitive Science and AI:
Fei-Fei Li sees a two-way relationship between AI and cognitive science, with AI serving as a tool for neuroscience and cognitive science. She finds inspiration in the brain’s incredible learning, creativity, empathy, and compassion. Specific areas of interest include energy-efficient algorithms, learning from exploration and curiosity, and mathematical formulations of empathy for AI.

Diversity in AI:
Fei-Fei Li co-founded AI for All in 2015 with Dr. Olga Ruzakovsky to address the lack of diversity in the AI field. The organization aims to increase the participation of underrepresented groups, create a more inclusive AI community, and address societal biases in AI systems. AI for All provides resources, workshops, and mentorship programs to support diversity and inclusion initiatives in AI.

The Crisis of Lack of Diversity and Representation:
The public’s concern about the potential risks of AI coexisted with the silent crisis of lack of diversity and representation in the AI field. In 2014, Fei-Fei Li was the only woman faculty in Stanford AI Lab, and the percentage of women and underrepresented racial minorities in the field was low.

The Connection between the Two Crises:
Li realized that addressing the diversity crisis was crucial to shaping the future of AI and preventing potential negative outcomes. The lack of diversity in AI leadership could lead to a narrow perspective in the development and deployment of AI technology.

Inception of AI for All:
Li, along with her student Olga and educator Rick Sommer, conceived the idea of AI for All in 2014. The program aimed to invite high school students to study and research AI at Stanford, exposing them to the technology’s human-centered mission and inspiring them to join the field.

The Motto of AI for All:
“AI will change the world. Who will change AI?” The belief that creating a diverse generation of AI leaders would lead to the development of human-centered technology and avoid dystopian scenarios.

Addressing the Diversity Crisis:
Raising awareness of the issue is an important first step. Addressing the pipeline issue by encouraging underrepresented and underserved communities to join the field of AI. Creating a welcoming and inclusive culture in workplaces and society for diverse AI professionals. Collaborating with civil society and government to incentivize diversity and inclusion programs.

Conclusion:
The diversity crisis in AI leadership is a perennial issue that requires ongoing efforts from all stakeholders. AI for All is an initiative that aims to address this issue by inspiring and supporting diverse students to join the field of AI.

Appreciation for Work:
Alex Wang expresses his appreciation for the impactful and important work being done in healthcare, diversity in AI, and human-centered artificial intelligence.

Well Wishes:
Alex extends his well wishes for the rest of the program and for the success of the business.

Closing Remarks:
Alex expresses his enjoyment in the conversation and concludes with a thank you.